Taxiing is one of the most important operational processes in airports. Accurately predicting aircraft taxi time is essential for achieving efficient and robust airport management. In recent decades, taxi time prediction has received considerable research attention and machine learning prediction algorithms are perhaps the most popular methods in this field. However, existing research mainly concentrates on developing various taxi time prediction methods by adopting existing machine learning models and testing their performance is different airports. Another vital process, i.e., feature selection, receives relatively less attention, which may lead to some unnecessary features being involved in the taxi time prediction process and eventually reducing the overall prediction accuracy. To address the shortage of traditional prediction methods, this study integrates a popular feature selection technique with different machine learning models to develop more efficient taxi time prediction methods. Before starting the prediction process, the RReliefF feature selection method is employed to identify important features and form the feature subset, which is helpful to not only improve the prediction accuracy but also reduce computational complexity. Based on the obtained feature subset, three prediction models including multiple linear regression, support vector regression and artificial neural network are trained and used to realize accurate taxi time prediction. To investigate the performance of the proposed prediction methods, historical flight data from an international airport is collected to test their performance. Experimental results indicate that feature selection technique has the potential to enhance the taxi time prediction ability and save computing resources during the prediction process.